A chess engine that plays like you.
Otter is trained on millions of human games to predict the move a real player would make — the plans, the patterns, the occasional slip — instead of always finding the objectively strongest one.
Plays like a person, not a calculator
Otter is conditioned on a target rating band and predicts moves the way a player at that level would plausibly make — including the blind spots typical of that band.
Rating-conditioned play
Pick a target Elo from 800 to 2600 and Otter matches that level's style and mistakes.
Time control & clock pressure
Considers both the time control format and remaining clock time to simulate how human decision quality degrades under dynamic time stress.
Move sequence memory
Conditioned on the game's move history, allowing the model to look at previous moves and judge opening choices, positional drift, and momentum.
Outperforms Maia 2
Achieves a 55.70% top-1 accuracy, outdoing Maia 2's reported 53.25% while using 34% fewer parameters and 33% less training data.